Illuminating humanist nature in teaching translation and interpreting studies: Devising an online customisable AI-driven subtitling course

The study addresses how to design a technology-enabled, learner-friendly online subtitling course that foregrounds the humanistic nature of translation within Translation and Interpreting Studies (TIS). In the context of rapid advances in AI and increased demand for online teaching post-COVID-19, the paper argues that technological integration must not come at the expense of human hermeneutic capacities. It poses two research questions: (1) Why are short video platforms and online T&I courses suitable and comparable case studies for analysis? (2) How can these comparisons inform the design of an AI-driven, learner-led online subtitling course? The purpose is to enhance student employability and practice in subtitling by combining human-centred pedagogy with AI-driven tools and customisable features, addressing the growing repository of audiovisual material and the need for high-quality, audience-tailored subtitles.

The paper adopts a theoretical framework combining polysystem theory, multimodality, and a humanistic approach to TIS. Polysystem theory (Even-Zohar; Díaz-Cintas) positions audiovisual translation within dynamic, heterogeneous cultural systems where linguistic, political, economic, and sociocultural factors shape translation. Multimodality (Baldry & Thibault; Kress & van Leeuwen) highlights the interplay of multiple semiotic resources (audio, visual, emojis, danmaku, gift-giving mechanisms) in meaning-making and informs course content that engages external technological and user-led modes. The humanistic perspective (Steiner; Ye) emphasizes preserving the hermeneutic ‘hard core’ of language users in translation pedagogy, viewing subtitling as an art of recreation and reformulation (Ma; Halliday) rather than mere word-for-word transfer. Prior pedagogical literature (Pym; Damson) supports integrating computer skills and CAT tools into translator training, while aligning with learner-centred, less stressful, and reflective practices (Elliott).

The study employs a mixed approach: (1) a student-led questionnaire and (2) case analyses of a short-video platform and two MOOCs. Questionnaire: Conducted with 36 first-year MTI postgraduate students at Sun Yat-Sen University (Autumn 2020) enrolled in a multimedia translation module. Three multiple-choice questions (plus an open prompt) probed the necessity of personalised subtitle settings, audience self-classification (junior/intermediate/senior), and preferred customisation options (e.g., completeness and language of subtitles, size, speed, fonts, colour, glossaries). Case analyses: (a) Douyin vs. Douyin Jisu editions—assessed for technological footprint (installation size, bandwidth, speed), economic incentives (bonuses, monetisation features), and UI (ad-free experience), extracting principles relevant to user-friendly, localised subtitling settings and creative economy integration. (b) MOOC Consecutive Interpreting—examined structure, interactivity, assessment criteria (quizzes 30%, discussions 10%, peer review 30%, final exam 30%), workload, and student demonstrations to inform a less stressful but interactive subtitling course design. (c) MOOC Working with Translation: Theories and Practice—reviewed global reach, engagement (video presentations, short quizzes, gamification), interactive commenting, and reflective discussion culture to adapt features that promote learner agency. Pedagogical design mapping: The study operationalises AI-driven subtitling across pre/during/post stages (adapting Wang’s three-stage translation technology): Pre (terminology building, formatting, genre rewriting, language selection, MT, presubtitling); During (face recognition login, progress tracking, workstation configuration, creative emoji insertion, feedback via creative economy, subtitle production/editing using OOONA, Wincaps, Yicat, etc.); Post (project management, post-editing/production/subtitling, localisation, file conversion, review). A 12-week syllabus integrates lectures and workshops (Aegisub, Trados, Yicat) with multimodal topics and assessments (attendance 10%; independent demonstration on a domestic platform with personalised settings 30%; group midterm on an international platform 30%; self-reflective report 30%).

• Questionnaire outcomes: 63.89% agreed that subtitle settings should be more personalised. Audience self-classification: 8.33% junior, 72.22% intermediate, 19.44% senior; over 80% identified as intermediate or senior. Participants preferred options enabling different subtitle types (bilingual, interlingual, intralingual), flexible completeness by language level, and inclusion of glossaries; speed and size adaptations were valued more than font/color changes. Learners expressed desire for platform-like options (e.g., Bilibili/Netflix/YouTube/Prime) to select English/bilingual/none.
• Case analysis—Douyin vs. Douyin Jisu: Jisu offers smaller install size/less bandwidth/faster operation; economic incentives (bonuses to users, multiple monetisation avenues); ad-free interface—features that can inform a subtitling course oriented to accessibility, efficiency, and creative economy alignment (e.g., monetisable uploads of subtitled short videos, shared rewards for creators/viewers).
• Case analysis—Consecutive Interpreting MOOC: Highly interactive with student demonstrations and structured assessments, but mark-driven tasks (mandatory comments) can add pressure. Findings support adopting interactive demonstrations while reducing stress via more flexible, reflective assessment.
• Case analysis—Working with Translation MOOC: High engagement through videos, limited compulsory quizzes, and reflective discussions (including emotive/emoji interactions) encourage learner autonomy—features to be adapted into the subtitling course.
• Pedagogical synthesis: A human-centred, AI-enabled course can combine personalised subtitle settings, multimodal awareness (e.g., emojis, danmaku), CAT tools, and platform familiarity, with less-stressful assessments to enhance learning outcomes and employability.

The findings address RQ1 by showing that short-video platforms (Douyin/Jisu) and MOOCs (Consecutive Interpreting, Working with Translation) are suitable comparative cases due to their popularity, robust interaction models, technological affordances, and economic/ecosystem features that parallel subtitling practice within a polysystemic and multimodal environment. For RQ2, the comparison informs concrete design choices: integrate AI-driven tooling across pre/during/post stages; enable personalised subtitle settings (language completeness, speed, size, glossaries); embed learner data features (face recognition login, progress tracking) to streamline experience; leverage platform conventions (short video formats, creative economy) to motivate authentic production; promote interactive student demonstrations while reducing assessment stress; and ensure content themes support humanistic, positive, and community-oriented outcomes. The humanistic lens ensures that technology augments, rather than replaces, hermeneutic judgement and aesthetic recreation in subtitling. Overall, the course operationalises theory (polysystem, multimodality, humanism) into a practical, employability-focused pedagogy that respects learner individuality and well-being.

The study proposes a 12-week, AI-driven, learner-led subtitling course that prioritises humanistic values while integrating contemporary technologies (CAT tools, face recognition login, big data updates). Based on questionnaire evidence and case analyses (Douyin/Jisu and two MOOCs), the course emphasises personalised subtitle settings, multimodal literacy, creative economy engagement, and less-stressful, reflective assessment. It contributes a theoretically grounded, practice-oriented blueprint adaptable to evolving audiovisual ecosystems. Future work could empirically evaluate learning outcomes, user experience, and employability impacts across broader learner populations and platforms, and refine AI features for personalisation and ethics in data use.

The paper does not report controlled learning-outcome evaluations; its questionnaire sample is limited to 36 MTI students from a single institution and semester, which may constrain generalisability. Case analyses are descriptive and context-specific (China-based platforms and selected MOOCs). Some preferred feature percentages beyond the main reported figures are not fully detailed due to figure constraints. Effectiveness of proposed AI-driven features (e.g., face recognition, progress tracking) and creative-economy integration is not empirically tested within the study.